US20030103003A1 - Method and device for co-operative radio direction-finding in transmission - Google Patents

Abstract

In a method of co-operative radio direction-finding in transmission in a system of radio direction-finding for a radiocommunications system comprising one or more transmission sources or transmitters, the transmitted signal comprises a reference sequence formed by a sequence of symbols {S_k}, the receiver comprising at least one array of several sensors coupled to a radio direction-finder. The method comprises the following steps:

a) determining the reference signal or signals d_i(t) associated with a transmitter m from the reference symbols {S_k} and from the modulation defined by the format of the signals; b) defining the instant of arrival t_k of one or more reference signals; c) from the signal x(t) received at the sensors, the instants of arrival t_k and the reference signal or signals d_i(t) derived from the step a), determining the value or values of incidence of the signals for a given transmitter. The method can be applied to GSM or UMTS signals.

Description

BACKGROUND OF THE INVENTION
• 1. Field of the Invention [0001]
• The present invention relates to a method and a device for co-operative radio direction-finding in transmission for signals comprising a reference signal. [0002]
• The invention can be applied especially to the case of signals comprising a synchronization signal or reference signal, consisting of a sequence of known symbols defined by the transmission standard used within the transmission system. [0003]
• For example, the GSM (Global System for Mobile communications), consisting of several reference signals, has this characteristic. In such a context, a reception system such as a base station may perform a selective direction-finding operation and associate a direction of arrival with a synchronization sequence. The techniques of selective direction-finding can be used to isolate one transmitter from the others on the basis of knowledge of its reference signal. It is thus possible to carry out a selective direction-finding operation for the transmitter, thus reducing the risk of ambiguity. [0004]
• 2. Description of the Prior Art [0005]
• The [0006] patent FR 2 764 074 describes a method for carrying out a direction-finding operation in the case of single-user operation and for linear modulations or waveforms that may be approximated in terms of linear modulations. In this patent, linearly modulated digital signals s_e(t) are sent. The symbols a₁, a₂, a₃, etc., associated with the signal, are transmitted at the period T_s and then filtered by a FIR (Finite Impulse Response) filter with a response h(t).
• Thus, the following is the relationship between the transmitted signal s[0007] _e(t) and the symbols a_n: $\begin{array}{cc}{s}_e\left(t\right)=\sum _nh\left(t-{\mathrm{nT}}_s\right)a_n& \left(1\right)\end{array}$

  
• where h(t) corresponds to the shaping filter. [0008]
• FIG. 1 gives a diagrammatic view of the principle of the method. [0009]
• The example given in this patent relates chiefly to the GSM standard in which the signal sent is a GMSK (Gaussian Minimum Shift Keying) phase modulation such that: [0010] $\begin{array}{cc}{s}_e\left(t\right)=\exp \{j\sum _n\Phi \left(t-{\mathrm{nT}}_s\right)a_n\}& \left(2\right)\end{array}$

  
• where a[0011] _n corresponds to a symbol sent at a rate of T_s samples per symbol.
• In order that the GMSK standard may form part of the linear modulations, the authors of this patent have made the following approximation: [0012] $\begin{array}{cc}{s}_e\left(t\right)\approx \sum _n{\mathrm{<figure-callout\; id="0"\; label="C"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">C</figure-callout>}}_0\left(t-{\mathrm{nT}}_s\right)j^nb_n\mathrm{with}b_n=\prod _{k=-\infty }^na_k& \left(3\right)\end{array}$

  
• The linearization is obtained, for example, by applying the method described in reference of Pierre Andre Laurent, “Exact and approximate construction of digital phase modulations by superposition of modulated pulse (AMP)” in IEEE Trans. on Communications Vol 34 (1986) pp 150-160. [0013]
• In the sequence of symbols {a[0014] _n}, there is a sequence of symbols {s_k} such that a_p+k−1={s_k} for 1≦k≦K. The sequence of symbols {s_k} is defined by the transmission standard and is called a learning sequence. In particular, the GSM standard comprises eight different sequences formed by K=26 symbols.
• The signal s[0015] _e(t) travels in transit towards the n^th antenna of a reception system through an FIR linear filter propagation channel C(t). Under these conditions, there is a relationship of linear filtering between the comb of symbols s(t) (such that s_e(t)=h(t)*s(t) where <<*>> designates the convolution product) and the sensor signals x_n(t), illustrated in FIG. 2.
• Given that C(t)=[C[0016] ₁(t) . . . C_N(t)]^T, the relationship between the transmitted symbols a_n and the received signal x(t) is expressed in the form: $\begin{array}{cc}\underset{\_}{x}\left(t\right)={C\left(t\right)}^{*}s_e\left(t\right)={\left(C^{*}h\left(t\right)\right)}^{*}s\left(t\right)=\sum _nG\left(t-{\mathrm{nT}}_s\right)a_n+{\underset{\_}{b}}_0\left(t\right)& \left(4\right)\end{array}$

  
• where [0017]
• for t=nT[0018] _s, s(t)=a_n and for t≠nT_s, s(t)=0.
• G(t)=C(t)*h(t) is a N×1-sized vector and [0019] b(t) is the vector formed by the background noise of the N receivers.
• The method disclosed in the [0020] patent FR 2 764 074 is subdivided into three steps:
• 1[0021] ^st step: Digitization of the signal received x(t) at the sampling period T_e. Since the input of the system is formed by symbols {a_n}, T_e must be chosen so that T_s=P T_e where P is an integer. Thus, if we write G_p−nP=G((p−nP)T_e) the relationship (4) becomes: $\begin{array}{cc}\underset{\_}{x}\left({\mathrm{pT}}_e\right)=\sum _nG_{p-\mathrm{nP}}a_n+{\underset{\_}{b}}_0\left(pT_e\right)& \left(5\right)\end{array}$

  
• In the description of the patent, the authors write P=2. The fact of choosing the sampling period T[0022] _e as a function of the symbol period T_s is a is heavy constraint. Generally, the equipment cannot be used to obtain T_s=P T_e (P integer) directly. The sampling is done at T′_e and, thereafter, to have a sampling at T_e, it is necessary to carry out digital processing operations which resample the signal at T_e. In the GSM standard, the signals sent verify the relationship 1/T_s=270 kHz. To meet the conditions of the Shannon theorem at reception, the system receives them at 1/T_e=500 kHz and T_s/T_e=500/270 is in fact a non-integer.
• 2[0023] ^nd step: Synchronization procedure: the goal is to determine the instant t₀=pT_e of appearance of the learning sequence {s_k} such that a_p+k−1={s_k} for 1≦k≦K: $\begin{array}{cc}\underset{\_}{x}\left(\left(k+p\right)T_e\right)=\sum _nG_{k-\mathrm{nP}}s_n+{\underset{\_}{b}}_0\left(\left(k+p\right)T_e\right)& \left(6\right)\end{array}$

  
• To this end, it is considered that the symbol comb s(n T[0024] _e) is the input signal and it is sought to maximize the following synchronization criterion in terms of τ:
• Cri(τ)={circumflex over (r)} _xs(τ)^H R _xx ⁻¹(τ){circumflex over (r)} _xs(τ)  (7)
• [0025] $\begin{array}{cc}\mathrm{With}{\hat{r}}_{\mathrm{xs}}\left(\tau \right)=\frac{1}{\mathrm{KP}}\sum _{k=1}^{\mathrm{KP}}\underset{\_}{x}\left(kT_e+\tau \right){s\left(kT_e\right)}^{*}\mathrm{and}R_{\mathrm{xx}}\left(\tau \right)=\frac{1}{\mathrm{KP}}\sum _{k=1}^{\mathrm{KP}}\underset{\_}{x}\left(kT_e+\tau \right){\underset{\_}{x}\left(kT_e+\tau \right)}^H& \left(7\right)\end{array}$

  
• K: number of symbols of the learning sequence [0026]
• and s(n T[0027] _e)=s_kP for n=kP and s(n T_e)=0 for n≠kP and where x ^H designates the transposition and conjugate of x.
• 3[0028] ^rd step: The signals x(t+t_k) and the symbol comb filters s(n T_e) associated with a source are used in order to isolate this source and carry out a direction-finding operation on a single source at the instant t₀=p T_e. With the use of the received signal x(t+t_k) and a single symbol comb filter s(n T_e), the method is said to be a single-user method.
• While the prior art method performs well, it nevertheless has certain drawbacks. It can be applied to waveforms that can be linearized or that can be approximated in such a form, and in single-user contexts. [0029]
• The method according to the invention relies on a novel approach that consists especially in exploiting the signal or reference sequence containing symbols proper to a standard used or to a given format for the signal, the received signal and the knowledge of the instance of arrival of the different reference sequences, to determine the coordinates of a source from a measurement of its incidence, this operation being commonly known as direction-finding. [0030]
• In the description, the terms “transmitters”, “users” or “useful transmitters” designate one and the same object. [0031]
SUMMARY OF THE INVENTION
• The invention relates to a method of co-operative radio direction-finding in transmission in a system of radio direction-finding for a radiocommunications system comprising one or more transmission sources or transmitters, the transmitted signal comprising a reference sequence formed by a sequence of symbols {s[0032] _k}, the receiver comprising at least one array of several sensors coupled to a radio direction-finder. The invention is characterized in that it comprises at least the following steps:
• a) determining the reference signal or signals d[0033] _i(t) associated with a transmitter m from the symbols and from the modulation defined by the format of the signals,
• b) defining the instant of arrival t[0034] _k of one or more reference signals,
• c) from the signal [0035] x(t) received at the sensors, the instants of arrival t_k and the reference signal or signals d_i(t) derived from the step a), determining the value or values of incidence of the signals for a given transmitter.
• The step a) is reiterated as many times as there are transmitters M[0036] ₀ in the system.
• It may comprise a step for the separation of the transmitters M[0037] ₀ through the estimation of the transfer matrix of the channel H^u for the transmission of all the useful signals and by the extraction, from this matrix, of the transfer matrices H_m ^u of the channel associated with a useful transmitter m in order to determine the corresponding observation vector x ^m(t).
• The step c) is carried out by means of a direction-finding operation by whitening, implemented on the observation vectors [0038] x ^m(t) or by projection.
• The step a) may be performed on a duration δt′>δt, where δt is the basic duration of the reference signal. [0039]
• A spatial filter W[0040] _m can be determined from the reference signal d_m(t) of a transmitter m in order to deduce the signal s_m (t) coming from the transmitter before applying a direction-finding step.
• The method can be applied especially in radio communications systems for GSM or UMTS type signals. [0041]
• The invention also relates to a radio direction-finding device comprising at least means adapted to implementing the steps of the above-mentioned method. [0042]
• The object of the invention has the following advantages in particular: [0043]
• the possibility of performing a direction-finding operation in a context that is wider (namely a single-user or multiple-user context and a single-path or multipath context) than the one offered by the prior art method. [0044]
• through the choice of the working time span or observation time span, the possibility of improving the results of the direction-finding, [0045]
• through the modification, for example the lengthening, of the duration of the reference signals, the possibility of improving the results of the direction-finding operation, [0046]
• the possibility of localizing mobiles to make directional transmission toward each mobile, on the basis of the knowledge of the direction of arrival of signals from the mobiles whose direction has been located, [0047]
• the possibility of localizing mobiles or base stations in a context of spectrum control, [0048]
• the reduction of the transmission power at constant range or the increasing of the transmission range at constant power, and consequently the reduction of interference with respect to the other cells, [0049]
• the method is applied for different waveforms, without specific processing, [0050]
• through the use of whitening or projection techniques that enable a more efficient rejection of interference, the possibility of improving the direction-finding results.[0051]
BRIEF DESCRIPTION OF THE DRAWINGS
• Other features and advantages of the invention shall appear from the following description, given by way of an illustration that in no way restricts the scope of the invention. The description is made with reference to the appended drawings, of which: [0052]
• FIGS. 1 and 2, show the principle of transmission and reception of a linear modulation according to the prior art, [0053]
• FIG. 3 shows a system of localization from a base station, [0054]
• FIG. 4 is a diagrammatic view of the principle of transmission-reception according to the invention, [0055]
• FIG. 5 shows the shape of a signal s[0056] _e(t) transmitted by a source m comprising a reference signal d(t),
• FIG. 6 shows the result of the sync operation in a multiple-user context, [0057]
• FIG. 7 is a block diagram showing the process of multiple-user detection and direction-finding, [0058]
• FIG. 8 is a drawing showing the lengthening of the duration of the reference signal, [0059]
• FIGS. 9 and 10 show two examples of implementation in the multiple-user case, [0060]
• FIGS. 11, 12, [0061] 13 and 14 show results of trials.
MORE DETAILED DESCRIPTION
• The following description, given by way of an illustration that in no way restricts the scope of the invention, relates to a radio direction-finding device whose goal especially is to locate the direction of the sources on an aircraft and/or on the ground. [0062]
• FIG. 3 gives schematic view of an exemplary system for the localizing of one or more mobiles Mi from a base station. The signals are propagated along direct paths (shown in solid lines in the figure) or along indirect paths (shown in dashes) as a result of the different reflections on obstacles. [0063]
• The [0064] base station 1 comprises for example an antenna array 2, behind which there is a system 3 for the reception of the signals coming from the antennas. A processor 4, adapted especially to the execution of the steps of the method according to the invention described in detail here below, is connected to the reception system
• FIG. 4 is a diagrammatic view of the principle of transmission-reception implemented in the method according to the invention. The elements identical to those of FIG. 2 bear the same references. The signal sent s[0065] _e(t) gets propagated by a transmission channel C(t) and is then received at the array of sensors of the base station.
• The transmitted signal s[0066] _e(t), an exemplary shape and structure of which are given in FIG. 5, comprises several reference signals d(t) that appear several times at given instants t₁, t₂ . . . . The reference signal d(t) is formed, for example, by a sequence of symbols {s_k} for 1≦k≦K, where K is the number of symbols of the sequence. The duration δt of the reference signal d(t) then depends on the number K, the symbol time T_s and the sampling period T_e.
• Step a) of the Method: Construction of the Reference Signal d(t) [0067]
• In the method, the reference signal d(t) is built from symbols of the learning sequence {s[0068] _k}, the symbol time T_s corresponding to the transmitted signal, the sampling period T_e and the modulation sent (the standard used for the signals, which may be the waveform h(t) in the case of the linear modulations).
• In the case of multiple users, with M[0069] ₀ users or useful transmitters, the step a) is reiterated as many times as there are users, in order to determine the reference signals d₁(t) . . . d_M0(t) for each of them. This makes it possible to obtain a multiple-user direction-finding by using the presence of the M₀ reference signals between the instants t_k and t_k+δt where δt is the duration of the reference signals.
• For example, for the GSM signals, the reference d(t) is computed by using the above-mentioned expression (2). [0070]
• For UMTS type signals, the method determines the reference sequence d(t) from spread-spectrum sequences known to those skilled in the art and from the expression (1) where h(t) is a Nyquist filter with a roll-off value of 0.25. [0071]
• This step a) is followed, for example, by a step for the digitization of the received signal [0072] x(t) at the sampling period T_e. In this step, it is not necessary to choose T_e such that T_s=PT_e where P is an integer.
• Step b) or Synchronization Step [0073]
• The next step consists in determining, in the received signal [0074] x(t), the instants t₁, t₂ etc of the appearance of the reference signal d(t) such that, according to the relationship (4): $\begin{array}{cc}\underset{\_}{x}\left(t+t_k\right)=\sum _nC\left(t\right)d\left(t-{\mathrm{nT}}_e\right)+{\underset{\_}{b}}_0\left(t+t_k\right)\mathrm{for}0<t<\delta t& \left(8\right)\end{array}$

  
• because [0075] x(t)=C(t)*s_e(t)+b ₀(t) and s_e(t+t_k)=d(t) for 0<t<δt. For this purpose, the method considers that the reference signal d(t) is the input signal and seeks to maximize the following synchronization criterion in terms of τ:
• Cri(τ)={circumflex over (r)} _xd(τ)^H R _xx ⁻¹(τ){circumflex over (r)} _xd(τ) $\begin{array}{cc}\mathrm{with}{\hat{r}}_{\mathrm{xd}}\left(\tau \right)=\frac{1}{\delta t}\sum _{t=O}^{\delta t}\underset{\_}{x}\left(t+\tau \right){d\left(t\right)}^{*}\mathrm{and}R_{\mathrm{xx}}\left(\tau \right)=\frac{1}{\delta t}\sum _{t=O}^{\delta t}\underset{\_}{x}\left(t+\tau \right){\underset{\_}{x}\left(t+\tau \right)}^H& \left(9\right)\end{array}$

  
• Step c) Determining of the Directions of Arrival or Direction-finding [0076]
• The signals [0077] x(t+t_k) and d(t) associated with a source m are used in order to isolate it and carry out a direction-finding operation on a single source at a given point in time t_k.
• The implementation of the step c) depends especially on the context of use, for example in the single-user or multiple-user framework. The description differentiates between these two cases. [0078]
Application of the Method in the Single-user Case
• In the method, first of all a search is made for the instants of synchronization at t[0079] _k=pT_e for example by using the method described in the patent FR 2 715 488. Given the different synchronization instants and the instants of appearance t_k of the learning sequence, the direction-finding is obtained, in the method, for example according to a MUSIC type method, known to those skilled in the art, on the reconstructed, modulated signal d(t) obtained from the step a) and the observations x(t+t_k) on the time interval 0≦t≦δt. A method such as this is described, for example, in R. O. Schmidt, “A signal subspace approach to multiple emitters location and spectral estimation”, PhD Thesis, Stanford University, CA, November 1981].
Application of the Method in the Multiple-user Case
• In a more general context, it may be desired to make a search, in the received signal [0080] x(t), for several sources which are associated with reference signals d_i(t) different from one another. In GSM, for example, the idea is to search for the sources associated with the eight possible TSC sequences, 1≦i≦8. Hence, the synchronization instants associated with the i^th synchronization d_i(t) will be called t_k(i).
Search for the Number M₀ of Useful Transmitters or Useful Units
• In the method, a search will be made initially for the number M[0081] ₀ of useful transmitters. To illustrate the process implemented, the description considers the case where there are two useful units or transmitters.
• FIG. 6 shows the presence of two reference signals d[0082] _i(t) and d_j(t) which are detected at the respective instants t_k(i) and t_k(j) and correspond to two users and two transmitters.
• If the difference Δt[0083] _ij=t_k(j)−t_k(i) is small enough, for example Δt_ij<αδt (typically α≈0.1), a multiple-user algorithm can be applied (for example by implementing a multiple-user direction-finding technique) by using the signals d_i(t) and d_j(t) in the model of the equation (16) between the instants t_k(i) and t_k(j)+δt, namely the time period or signal section where the two users are present. According to the expression (6), the signal x(t+t_k(i)) is written as follows: $\begin{array}{cc}\begin{array}{c}x\left(t+t_k\left(i\right)\right)=\sum _{n=L/2}^{L/2}C_i\left(t\right)d_i\left(t-\mathrm{nTe}\right)+\\ \sum _{n=L/2}^{L/2}C_j\left(t+\Delta t_{\mathrm{ij}}\right)d_j\left(t+\Delta t_{\mathrm{ij}}-{\mathrm{nT}}_e\right)+{\mathrm{<figure-callout\; id="2"\; label="b"\; filenames="US20030103003A1-20030605-D00002.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">b</figure-callout>}}^2\left(t\right)\end{array}\mathrm{for}0<t<\delta t_{\mathrm{ij}}& \left(10\right)\end{array}$

  
• where δt[0084] _ij corresponds to the signal section used for the processing, in multiple-user mode, of the i^th and the j^th signals.
• Where [0085] b ²(t) is a noise formed by the background noise b ₀(t) and interferers having no synchronization signal, close enough to the instants t_k(i) and t_k(j). This expression can also be expressed in the following form (11)

  
• Where H[0086] _m ^u represents the matrix of the propagation channel for the useful transmitter m and r ^m _L(t) is its reference vector.
• For M[0087] ₀=2, the values H_m ^u and the values r ^m _L(t) verify: $\begin{array}{cc}{\underset{\_}{r}}_{\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">L</figure-callout>}}^1\left(t\right)=\left[\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">d</figure-callout>}}_1(t-\left(L/2\right)\mathrm{Te})\\ ·\\ {\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">d</figure-callout>}}_1(t+\left(L/2\right)\mathrm{Te})\end{array}\right]\mathrm{and}{\underset{\_}{r}}_{\mathrm{<figure-callout\; id="2"\; label="L"\; filenames="US20030103003A1-20030605-D00002.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">L</figure-callout>}}^2\left(t\right)=\left[\begin{array}{c}{d}_2(t+\Delta t_{12}-\left(L/2\right)\mathrm{Te}\\ ·\\ d_2(t+\Delta {\mathrm{<figure-callout\; id="12"\; label="t"\; filenames="US20030103003A1-20030605-D00004.png"\; state="\{\{state\}\}">t</figure-callout>}}_{12}+\left(L/2\right)\mathrm{Te}\end{array}\right]& \left(12\right)\end{array}$

  

   and
• H ₁ ^u =[C ₁(t−(L/2)T _e) . . . C ₁(t+(L/2)T _e)]
• H ₂ ^u =[C ₂(t−(L/2)T _e +Δt ₁₂) . . . C ₂(t+(L/2)T _e +Δt ₁₂)]  (13)
• Where L is the number of time shifts of the reference signal. The optimal procedure, for example, would be to choose L so that it is equal to the temporal spread of the channel of the transmitter with the greatest spread. [0088]
• More generally, the number of useful transmitters M[0089] ₀ is equal to the number of instants t_k(i) belonging to the time interval [t₀t₀+Δt_ij [,where Δt_ij=αδt. The value of α is typically α≈0.1 and δt is the duration of the synchronization signals.
• The process of multiple-user direction-finding and detection is then illustrated in FIG. 7. [0090]
• To put it briefly, the method in the case of a multiple-user context: [0091]
• determines the different instants of synchronization t[0092] _k(i) for all the signals i,
• determines the number of useful units or transmitters M[0093] ₀,
• carries out the direction-finding operation for each useful unit (transmitter) M[0094] ₀ in knowing the associated reference signal d_i(t) and the instants of synchronization t_k(i).
• The direction-finding step is performed for example according to different methods, some of which are given here below by way of an indication that in no way restricts the scope of the invention. Before the direction-finding operation is implemented, the method entails an estimation of the different propagation channels and the separation of the useful transmitters. [0095]
• For this purpose, the method considers the interferers correlated with the useful transmitter to be useful transmitters. The context is thus a multiple-user context with M[0096] ₀ useful transmitters and M-M₀ interference transmitters .
• After the separation of the useful transmitters, the expression of the signal [0097] x ^m(t) received at the array of sensors and coming from the user is equivalent to that of the single-user mode and a single-user direction-finding algorithm can then be applied to this signal.
• Thus, the signal received at the sensors is expressed as follows: [0098] $\begin{array}{cc}\underset{\_}{x}\left(t\right)=\sum _{m=1}^{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0}H_m^u{\underset{\_}{r}}_L^m\left(t\right)+{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)\mathrm{Where}{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)=\sum _{m^{\prime }={\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0+1}^MH_{m^{\prime }}{\underset{\_}{s}}_L^{m^{\prime }}\left(t\right)+b\left(t\right)& \left(14\right)\end{array}$

  
• The matrices representing the transmission channel H[0099] _m ^u and H_m′, have the following expression (15): $H_m^u=\sum _{p=1}^{P_m}{\rho }_{\mathrm{mp}}\underset{\_}{a}\left(u_{\mathrm{mp}}\right)h_{\mathrm{mp}}^{\mathrm{uT}}=A_mB_m^u\mathrm{and}$ $H_{m^{\prime }}=\sum _{p=1}^{P_{m^{\prime }}}{\rho }_{m^{\prime }p}\underset{\_}{a}\left(u_{m^{\prime }p}\right)h_{m^{\prime }p}^T=A_{m^{\prime }}B_{m^{\prime }}$

  
• Where A[0100] _m=[a(u_m1) . . . a(u_mPm)] is the matrix of the direction vectors of the multipaths of the m^th transmitter, B_m ^uT=[_ρm1h_m1 ^u . . . _ρmPmh_mPm ^u] and B_m′ ^T=[_ρm′1h_m′1 . . . _m′Pm′h_m′Pm′], u_mp is the incidence of the p^th path of the m^th user. The expression (14) of x(t) can be written as follows:
• x (t)=H _u r ^u _L(t)+ b ^m0(t)  (16)
• where H[0101] _u represents the matrix of the propagation channels of the M₀ useful transmitters and r ^u _L(t) represents the associated multiple-user reference vector. $\mathrm{with}H_u=\left[H_1^u\dots H_{\mathrm{M0}}^u\right]$ $\mathrm{and}{\underset{\_}{r}}_L^u\left(t\right)=\left[\begin{array}{c}{\underset{\_}{r}}_{\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">L</figure-callout>}}^1\left(t\right)\\ ·\\ {\underset{\_}{r}}_{\mathrm{<figure-callout\; id="0"\; label="L\; M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">L</figure-callout>}}^{M_{}}\left(t\right)\end{array}\right]=\left[\begin{array}{c}{\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">d</figure-callout>}}_1(t-\left(L/2\right)\mathrm{Te})\\ ·\\ {\mathrm{<figure-callout\; id="1"\; label="d"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">d</figure-callout>}}_1(t+\left(L/2\right)\mathrm{Te})\\ ·\\ d_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0}(t-\left(L/2\right)\mathrm{Te})\\ ·\\ d_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0}(t+\left(L/2\right)\mathrm{Te})\end{array}\right]$

  
• The exponent u refers to the transmitters known as useful transmitters and L corresponds to the number of time shifts of the reference signal. The optimal way would be to choose L so that it is equal to the temporal spread of the channel of the transmitter with the greatest spread. [0102]
• The goal of the multiple-user technique consists firstly in separating the M[0103] ₀ useful signals to then apply a single-user type technique to each of the useful transmitters.
Separation of the Transmitters or Useful Transmitters
• The separation of the transmitters consists in estimating the transfer matrix H[0104] _u=[H₁ ^u . . . H_M0 ^u] and then extracting the matrices H_m ^u, associated with each of the M₀ useful transmitters, from H_u. The method estimates the matrix H_u according to the least error squares method as in the single-user case. Thus, according to the expression (16), the method estimates H_u in minimizing the difference between H_u r ^u _L(t) and x(t). By obtaining this least error square value between the instants t_min and t_max we get:
• Ĥ _u ={circumflex over (R)} _xr ^u({circumflex over (R)} _rr ^u)⁻¹  (17)
• [0105] $\mathrm{with}{\hat{R}}_{\mathrm{xr}}^u=\frac{1}{t_{\max }-t_{\min }}\sum _{t=t_{\min }}^{t_{\max }}\underset{\_}{x}\left(t\right){{\underset{\_}{r}}_L^u\left(t\right)}^H$ $\mathrm{and}{\hat{R}}_{\mathrm{rr}}^u=\frac{1}{t_{\max }-t_{\min }}\sum _{t=t_{\min }}^{t_{\max }}{\underset{\_}{r}}_L^u\left(t\right){\left({\underset{\_}{r}}_L^u\left(t\right)\right)}^H$

  
• Writing [0106] b ^M0(t)=x(t)−H_u r ^u _L(t) it is deduced from this that the covariance matrix of this noise+interferers vector is equal to: $\begin{array}{c}{\hat{R}}_{\mathrm{bb}}^u=\frac{1}{t_{\max }-t_{\min }}\sum _{t=t_{\min }}^{t_{\max }}{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right){{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)}^H\\ ={\hat{R}}_{\mathrm{xx}}-{\hat{H}}_u{\hat{R}}_{\mathrm{xr}}^{\mathrm{uH}}={\hat{R}}_{\mathrm{xx}}-{{\hat{R}}_{\mathrm{xr}}^u\left({\hat{R}}_{\mathrm{rr}}^u\right)}^{-1}{\hat{R}}_{\mathrm{xr}}^{\mathrm{uH}}\end{array}$ $\mathrm{with}{\hat{R}}_{\mathrm{xx}}=\frac{1}{t_{\max }-t_{\min }}\sum _{t=t_{\min }}^{t_{\max }}\underset{\_}{x}\left(t\right){\underset{\_}{x}\left(t\right)}^H$

  

   {circumflex over (R)} _bb ^u ={circumflex over (R)} _xx− {circumflex over (R)} _xr ^u({circumflex over (R)} _rr ^u)⁻¹ {circumflex over (R)} _xr ^{u _†}   (18)
• Since the matrices Ĥ[0107] _m ^u are sub-blocks of the matrix Ĥ_u, when Ĥ_u=[Ĥ₁ ^u . . . Ĥ_M0 ^u] is verified, they are extracted directly from Ĥ_u by the selection of the right columns. Thus, the useful signals have been separated and the covariance matrix {circumflex over (R)}_bb ^u of the noise+interferers component has been estimated. It is observed that, to estimate the matrices {circumflex over (R)}_bb ^u and Ĥ_m ^u, the matrix {circumflex over (R)}_rr ^u, which is the covariance matrix of the vector r ^u _L(t), is brought into play. This matrix, according to (16), has the following structure: $\begin{array}{cc}{\hat{R}}_{\mathrm{rr}}^u=\left[\begin{array}{ccccccc}{\hat{r}}_{11}\left(0\right)& ·& {{\hat{r}}_{11}\left(L\right)}^{*}& & ·& & {{\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_01}\left(L\right)}^{*}\\ ·& ·& & & & & \\ {\hat{r}}_{11}\left(L\right)& & {\hat{r}}_{11}\left(0\right)& & & & \\ ·& & & ·& & & ·\\ {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_01}\left(0\right)& ·& {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_01}\left(-L\right)& & {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0}\left(0\right)& & \\ ·& & & & & ·& \\ {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_01}\left(L\right)& ·& {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_01}\left(0\right)& ·& {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0}\left(L_T\right)& ·& {\hat{r}}_{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0M_0}\end{array}\right]& \left(19\right)\end{array}$

  
• It can be seen that the matrix brings into play intercorrelation terms {circumflex over (r)}[0108] _ij(T) between the reference signals d_i(t) and d_j(t−T) such that: $\begin{array}{cc}{\hat{r}}_{\mathrm{ij}}=\frac{1}{t_{\max }-t_{\min }}\sum _{t=t_{\min }}^{t_{\max }}d_i\left(t\right){d_j\left(t-T\right)}^{*}& \left(20\right)\end{array}$

  
• Thus, the multiple-user method makes use of the coefficients of correlations {circumflex over (r)}[0109] _ij(T) between the reference signals which are not necessarily zero. This point means that it is possible to intuitively verify that the use of the coefficients {circumflex over (r)}_ij(T) is necessary to separate the reference signals d_i(t) and d_j(t) in the correlated case.
• From the knowledge of the transfer matrix H[0110] _m ^u of the m^th useful transmitter, it is possible to build up an observation vector x ^m(t), the other useful transmitters having been rejected. The following transformation is carried out according to (14): $\begin{array}{cc}\begin{array}{c}{\underset{\_}{x}}^m\left(t\right)=\underset{\_}{x}\left(t\right)-\sum _{m^{\prime }\ne m}H_{m^{\prime }}^u{\underset{\_}{r}}_L^{m^{\prime }}\left(t\right)\\ \Downarrow \\ {\underset{\_}{x}}^m\left(t\right)=H_m^u{\underset{\_}{r}}_L^m\left(t\right)+{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)\end{array}& \left(21\right)\end{array}$

  
• The expression of the signal [0111] x ^m(t) is equivalent to that of the single-user case where M₀=1 and the method applies a single-user direction-finding algorithm to the signal x ^m(t). Two examples of this algorithm are given here below by way of an illustration that in no way restricts the scope of the invention.
Direction-finding by Whitening
• It being known that the transfer matrix is equal to H[0112] _m ^u=A_mB_m ^u and assuming that r ^u _L(t) and b ^M0(t) are decorrelated, the covariance matrix R_xx ^m of the signal x ^m(t) has the following structure according to the expressions (21) and (15):
• R _xx ^m =E[x ^m(t) x ^m(t)^H ]=A _m R _sm ^u A _m ^H +R _bb ^u  (22)
• with
• R _sm ^u =B _m ^u E[r ^m _L(t) r ^m _L(t)^H ]B _m ^uH
• Where R[0113] _bb ^u is the covariance matrix of the signal b ^M0(t) whose expression is that of the formula (18). The matrix A_m is constituted by the direction vectors a(u_mp) of the p^th path of the m^th useful transmitter. Since R_bb ^u is different from σ²I and since this matrix is known, the MUSIC method is applied by whitening the observations x ^m(t) by the matrix R_bb ^u. According to (21) the matrix R_xx ^m can be written as follows:
• R _xx ^m =H _m ^u R _rr ^m H _m ^uH +R _bb ^u with R _rr ^m =E[r ^m _L(t) r ^m _L(t)^H]  (23)
• Under these conditions the following are the steps of the algorithm: [0114]
• 1[0115] ^st step: Computation of {circumflex over (R)}_xx=E[x(t) x(t)^H], {circumflex over (R)}_xr ^u=E[x(t) r ^u _L(t)^H] and {circumflex over (R)}_rr ^u=E[r ^u _L(t) r ^u _L(t)^H]
• 2[0116] ^nd step: Computation of the transfer matrix Ĥ_u={circumflex over (R)}_xr ^u({circumflex over (R)}_rr ^u)⁻¹
• 3[0117] ^rd step: Extraction of the estimated matrices of the propagation channels for each of the transmitters m Ĥ_m ^u for 1≦m≦M₀ of the matrix Ĥ_u, given that Ĥ_u=[Ĥ₁ ^u . . . Ĥ_M0 ^u]
• 4[0118] ^th step: Computation of the noise matrix {circumflex over (R)}_bb ^u={circumflex over (R)}xx−{circumflex over (R)}_xr ^u({circumflex over (R)}_rr ^u)⁻¹{circumflex over (R)}_xr ^uH For each transmitter or useful transmitter m with a transfer matrix Ĥ_m ^u, initialization is performed by m=1:
• 5[0119] ^th step: Computation of {circumflex over (R)}_xx ^m=Ĥ_m ^u{circumflex over (R)}_rr ^mĤ_m ^uH+{circumflex over (R)}_bb ^u with {circumflex over (R)}_rr ^m=E[r ^m _L(t) r ^m _L(t)^H]
• 6[0120] ^th step: Whitening of {circumflex over (R)}_xx ^m by {circumflex over (R)}_bb ^u such that: R_xx ^bm=W^½{circumflex over (R)}_xx ^mW^−½H with {circumflex over (R)}_bb ^u=W^½W^½H
• 7[0121] ^th step: Application of MUSIC on R_xx ^bm with the direction vectors a ^b(u)=W^−½ a(u) to estimate the incidences u_mp for 1≦p≦P_m of the multipaths of the m^th transmitter.
• 8[0122] ^th step: m=m+1
• 9[0123] ^th step: if m≦M₀ return to the 5^th step.
• At output, the method possesses the incidence values u[0124] _ip(t_k(i)) associated with the k^th instants of the p^th path of the i^th reference signal of this source. The incidence values u_ip(t) may be either azimuth values θ_ip(t) for a 1D direction-finding operation or they may be the pair of parameters (θ_ip(t), Δ_ip(t)) for a 2D direction-finding operation where Δ_ip(t) is the elevation angle.
Direction-finding by Projection
• As in the single-user case, the method uses, for example, a method by projection. Indeed, given that H[0125] _m ^u=A_mB_m ^u, the expression of x ^m(t) is as follows according to (21):
• x ^m(t)=A _m B _m ^u r ^m _L(t)+b ^m0(t)  (24)
• where the matrix A[0126] _m=[a(u_m1) . . . a(u_mPm)] is constituted by the direction vectors a(u_mp) of the multipaths of the m^th useful transmitter. Given, according to (15) that H_m′=A_m′B_m′, the expression of the interference signals b ^M0(t) of the expression (14) becomes $\begin{array}{cc}{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)=\sum _{m^{\prime }={\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0+1}^MA_m·B_m·{\underset{\_}{s}}_L^{m^{\prime }}\left(t\right)+b\left(t\right)=A_b^{\mathrm{M0}}{\underset{\_}{s}}_b\left(t\right)+b\left(t\right)& \left(25\right)\end{array}$

  
• where A[0127] _b ^M0=[A_M0+1 . . . A_M] is the matrix of the direction vectors of all the interference sources and s _b(t)=[{B_M0+1 s ^M0+1 _L(t)}^T . . . {B_M s ^M _L(t)}^T]^T is the vector of the interference signals.
• In the specific example of the use of the method in a multiple-user case, the steps of the algorithm for direction-finding by projection are the following: [0128]
• 1[0129] ^st step: Computation of {circumflex over (R)}_xx=E[x(t) x(t)^H], {circumflex over (R)}_xr ^u=E[x(t) r ^u _L(t)^H] and {circumflex over (R)}_rr ^u=E[r ^u _L(t) r ^u _L(t)^H]
• 2[0130] ^nd step: Computation of the transfer matrix of the propagation channel Ĥ_u={circumflex over (R)}_xr ^u({circumflex over (R)}_rr ^u)⁻¹
• 3[0131] ^rd step: Extraction of the estimated matrices for the different propagation channels corresponding to the users or useful transmitters m Ĥ_m ^u for 1≦m≦M₀ of the matrix Ĥ_u it being known Ĥ_u=[Ĥ₁ ^u . . . Ĥ_M0 ^u]
• 4[0132] ^th step: Computation of the noise matrix {circumflex over (R)}_bb ^u={circumflex over (R)}_xx−{circumflex over (R)}_xr ^u({circumflex over (R)}_rr ^u)⁻¹{circumflex over (R)}_xr ^uH
• 5[0133] ^th step: Application of MUSIC to {circumflex over (R)}_bb ^u and deduction of A_b ^M0
• 6[0134] ^th step: Computation of Π_b=I−A_b ^M0(A_b ^M0HA_b ^M0)⁻¹ A_b ^M0H
• For each Transmitter m with a Transfer Matrix Ĥ[0135] _m ^u. Initialization by m=1:
• 7[0136] ^th step: Computation of {circumflex over (R)}_xx ^M=Ĥ_n ^u{circumflex over (R)}_rr ^mĤ_m ^uH with {circumflex over (R)}_rr ^m=E[r ^m _L(t) r ^m _L(t)^H]
• 8[0137] ^th step: Computation of the matrix R_xx ^um=Π_b ({circumflex over (R)}_xx ^m)Π_b
• 9[0138] ^th step: Application of MUSIC to R_xx ^um with the direction vectors a _pro(u)=Π_b a(u) to estimate the incidence values u_mp pour 1≦P≦p_m of the multipaths of the m^th transmitter.
• 10[0139] ^th step: m=m+1
• 11[0140] ^th step: if m≦M₀ return to the 7^th step.
• At the end of this step, the method possesses the incidence values u[0141] _ip(t_k(i)) associated with the k^th instants of the p^th path of the i^th reference signal of this source. The incidence values u_ip(t) may be either azimuth values θ_ip(t) for a 1D direction-finding operation or they may be the pair of parameters (θ_ip(t), Δ_ip(t)) for a 2D direction-finding operation where Δ_ip(t) is the elevation angle.
Improvement of the Direction-finding Results
• The steps that have been described here above may be improved. Generally, the direction-finding performance values are inversely proportional to the time duration δt of the reference signal. Thus, for a time duration of the learning signal δt′ equal to αδt, the precision of the direction-finding is {square root}{square root over (α)} more precise. [0142]
• The method then comprises a step in which the duration of the reference signal is increased. As can be seen in FIG. 8, the method consists of the deduction, from the observed signal [0143] x(t), of a reference signal s_m(t) with a greater duration δt′ such that δt′>δt.
• This value δt′ is determined for example as a function of the standard used. [0144]
• For example, in the GSM standard, the eight TSC signals have a duration δt=96 μs and the duration of a burst from a user is δt′=577 μs: At best, it is possible to have a ratio α=δt′/δt=6 and thus have more precise direction-finding values {square root}{square root over (α)}=2.5. [0145]
• To obtain this result, this method implements the following steps: [0146]
Determining the Reference Sequence
• Step a) is executed in considering a time duration δt′>δt in order to obtain one or more reference signals d[0147] _m(t) for one or more useful transmitters m.
Determining a Spatial Filter
• From the reference signal d[0148] _m(t) with a specified duration δt determined at the step a) and from the signal x(t) received by the sensors, the method determines a spatial filter Wm such that d_m(t)=W_m x(t)
Determining the Signal s_m(t)
• To determine the signal s[0149] _m(t) with the duration δt′, first of all a spatial filtering is conducted on the basis of the learning sequence d_m(t) to obtain an estimate ŝ_m(t) of the signal s_m(t) with a duration δt′. Given that ŝ_m(t) is not exactly equal to s_m(t), it is possible to envisage its conversion into a signal {circumflex over ({circumflex over (s)})}_m(t) close to s_m(t).
• Depending on the characteristics of the received signal, the method makes the following improvements: [0150]
• (a) Demodulation of ŝ[0151] _m(t) for the digital signals: Given that, in most cases, the characteristics of the intercepted digital signals are known, for example for the GSM, the UMTS, one alternative is to demodulate the signal ŝ_m(t) and then remodulate the symbols detected and obtain the signal {circumflex over ({circumflex over (s)})}_m(t).
• (b) Case of the signals with constant modulus: Given that these signals have a constant modulus equal to 1, |s[0152] _m(t)|=1, the method builds a reference signal {circumflex over ({circumflex over (s)})}_m(t) as follows: {circumflex over ({circumflex over (s)})}_m(t)=exp(j angle(ŝ_m(t)) where angle(z) designates the argument of the complex z. According to the expression (2), this principle can be applied in GSM mode because the symbols are modulated in a phase modulation which is the GMSK.
• (c) Case of any signals: In these conditions, the usable reference sequence can only be {circumflex over ({circumflex over (s)})}[0153] _m(t) such that {circumflex over ({circumflex over (s)})}_m(t)=ŝ_m(t)
Determining of the Spatial Filter of the Signal s_m(t)
• In the presence of M[0154] ₀ useful signals, the signal x(t) received at the N sensors of the array is written as follows (26): $\underset{\_}{x}\left(t\right)=\sum _{m=1}^{{\mathrm{<figure-callout\; id="0"\; label="M"\; filenames="US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">M</figure-callout>}}_0}H_m^u{\underset{\_}{r}}_L^m\left(t\right)+{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)=H^u{\underset{\_}{r}}_L^u\left(t\right)+{\underset{\_}{b}}^{\mathrm{M0}}\left(t\right)\mathrm{for}{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">t</figure-callout>}}_1\le t\le {\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">t</figure-callout>}}_1+\delta t$ $\mathrm{where}{\underset{\_}{r}}_L^u\left(t\right)=\left[\begin{array}{c}{\underset{\_}{r}}_L^I\left(t\right)\\ .\\ {\underset{\_}{r}}_L^{\mathrm{M0}}\left(t\right)\end{array}\right]\mathrm{and}{\underset{\_}{r}}_L^m\left(t\right)=\left[\begin{array}{c}{d}_m\left(t-\left(L/2\right)\mathrm{Te}\right)\\ .\\ d_m\left(t+\left(L/2\right)\mathrm{Te}\right)\end{array}\right]$

  
• The signal d[0155] _m(t) is the learning signal of the m^th transmitter for t₁≦t≦t₁+δt. The reference vector r ^u _L(t) therefore consists of the reference signals d_m(t−k) of the M₀ useful signals. Assuming that the m^th transmitter sends a signal s_m(t) as in FIG. 8, the following is deduced from this, for all the instants t:
• x (t)=H _u s ^u _L(t)+ b ^M0(t)  (27)
• [0156] $\mathrm{where}{\underset{\_}{s}}_L^u\left(t\right)=\left[\begin{array}{c}{\underset{\_}{s}}_{\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">L</figure-callout>}}^1\left(t\right)\\ .\\ {\underset{\_}{s}}_L^{\mathrm{M0}}\left(t\right)\end{array}\right]\mathrm{and}{\underset{\_}{s}}_L^m\left(t\right)=\left[\begin{array}{c}{s}_m\left(t-\left(L/2\right)\mathrm{Te}\right)\\ .\\ s_m\left(t+\left(L/2\right)\mathrm{Te}\right)\end{array}\right]$

  
• From the relationship (26), the method determines the space-time filter W enabling the estimation of [0157] r ^u _L(t) from x(t). According to (27), this same filter is used to estimate the signal s ^u _L(t) from x(t). In the absence of noise and interferers, W verifies: s ^u _L(t)=W x(t). Thus, W is estimated from the equation (26) in minimizing the difference between W x(t) and r ^u _L(t) for t₁≦t≦t₁+δt:
• Ŵ={circumflex over (R)} _ry ^u({circumflex over (R)} _xx ^u)⁻¹  (28) $\mathrm{with}{\hat{R}}_{\mathrm{rx}}^u=\frac{1}{\delta t}\sum _{t=t_1}^{{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">t</figure-callout>}}_1+\delta t}{\underset{\_}{r}}_L^u\left(t\right){\underset{\_}{x}\left(t\right)}^H\mathrm{and}{\hat{R}}_{\mathrm{xx}}^u=\frac{1}{\delta t}\sum _{t=t_1}^{{\mathrm{<figure-callout\; id="1"\; label="t"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">t</figure-callout>}}_1+\delta t}\underset{\_}{x}\left(t\right){\underset{\_}{x}\left(t\right)}^H$

  
• From this value, an estimate [0158] ŝ ^u _L(t) of the reference vector s ^u _L(t) is then deduced for t₁′≦t≦t₁′+δt′ verifying:
• ŝ ^u _L(t)=Ŵx (t) for t ₁ ′≦t≦t ₁ ′+δt′(29) $\mathrm{where}{\widehat{\underset{\_}{s}}}_L^u\left(t\right)=\left[\begin{array}{c}{\underset{\_}{\hat{s}}}_{\mathrm{<figure-callout\; id="1"\; label="L"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">L</figure-callout>}}^1\left(t\right)\\ .\\ {\underset{\_}{\hat{s}}}_L^{\mathrm{M0}}\left(t\right)\end{array}\right]\mathrm{and}{\underset{\_}{\hat{s}}}_L^m\left(t\right)=\left[\begin{array}{c}{\hat{s}}_m\left(t-\left(L/2\right)\mathrm{Te}\right)\\ .\\ {\hat{s}}_m\left(t+\left(L/2\right)\mathrm{Te}\right)\end{array}\right]$

  
• The signal ŝ[0159] _m(t) is the estimated reference sequence of the m^th transmitter on a duration δt′. As indicated at the beginning of this section, this estimation ŝ_m(t) could be refined by being converted into a signal {circumflex over ({circumflex over (s)})}_m(t).
Conversion of a Multiple-source Case into a Multiple-user Case
• FIG. 9 represents the time range in which the multiple-user algorithm is applicable. This corresponds to a duration δt[0160] ₁₂ smaller than the duration δt of the sequences demarcated by the instants t₁₂ and t₁₂+δt₁₂.
• To be able to apply the multiple-user algorithm without reducing the duration of the reference signals, it is therefore necessary to estimate the sequences {circumflex over ({circumflex over (s)})}[0161] ₁(t) and {circumflex over ({circumflex over (s)})}₂(t) of FIG. 10.
• First of all, the signals ŝ[0162] _m(t) will be estimated in single-user mode for 1≦m≦M₀ from the observations x(t) and the reference signal d_m(t) such that r ^u _L(t)=r ^m _L(t). The signals ŝ_m(t) are therefore estimated independently. As indicated here above, the M₀ estimations could be refined by being converted into signals {circumflex over ({circumflex over (s)})}_m(t).
EXAMPLES OF IMPLEMENTATION OF THE METHOD ACCORDING TO THE INVENTOIN
• FIG. 11 shows the results obtained with two plane arrays of N=5 sensors, comprising one circular array with a radius of R=21.1 cm and one “V” shaped array . The azimuth θ is the incidence of a source in the plane of the array and the elevation angle Δ is the incidence with respect to the perpendicular of the plane of the array. For the simulations in (θ, Δ), the direction-finding is done according to the parameters (u,v) which are the components of the wave vector of the source in the plane of the array. This components may be expressed in the form:[0163]
• u=cos(Δ) cos(θ) and v=cos(Δ) sin(θ)  (30)
• The two arrays are adapted to the frequency range 900 MHz-1 GHz of the GSM. In the simulations, the center frequency is equal to f[0164] _ant=900 Mhz and thus the wavelength is equal to λ=c/f_ant=33.33 cm.
• The direction-finding performance characteristics will then be estimated, either with respect to the parameter θ or with respect to the parameters (u,v) depending on whether it is a 1D or a 2D direction-finding operation that is performed. These performance characteristics are computed in terms of MSE (mean square error) of the parameter θ[0165] _m or (u_m,v_m) for the m^th source. For this purpose, several operations of direction-finding are carried out to find the direction of the source having the incidence θ_m or (u_m,v_m) to estimate the incidence {circumflex over (θ)}_m(k) or (û_m(k), {circumflex over (v)}_m(k)) at the k^th iteration. At each iteration, the same signal configuration is kept while, at the same time, the noise and the symbols of the incident signals are drawn randomly. Consequently, for a source with an incidence θ_m or (u_m,v_m), the computations of the MSE in 1D and 2D direction-finding are the following:
• 1D direction-finding in θ: [0166] ${\mathrm{\Delta \theta }}_{\mathrm{EQM}}^m=\sqrt{\frac{1}{\mathrm{nb}}\sum _{k=1}^{\mathrm{nb}}{\left({\hat{\theta }}_m\left(k\right)-{\theta }_m\right)}^2}$

  
• 2D direction-finding in (u,v): [0167] $\Delta R_{\mathrm{EQM}}^m=\sqrt{\frac{1}{\mathrm{nb}}\sum _{k=1}^{\mathrm{nb}}{\left({\hat{u}}_m\left(k\right)-u_m\right)}^2+{\left({\hat{v}}_m\left(k\right)-v_m\right)}^2}$

  
• where nb designates the number of iterations to compute the MES values: the MES of the m[0168] ^th source is parametrized by Δθ_EQM ^m for the 1D direction-finding and by ΔR_EQM ^m for the 2D direction-finding.
• Finally, a spatial distance must be set up between a useful transmitter with an incidence θ[0169] _m or (u_m,v_m) and interferers with an incidence θ_m or (u_m,v_m) for 2≦m≦M. This distance depends on the direction vectors of the sources whose expression is given, for example, in R. O. Schmidt, “A signal subspace approach to multiple emitters location and spectral estimation”, PhD Thesis, Stanford University, CA, November 1981. Taking a _m to denote the direction vector of the m^th source, this distance is defined by the projection of the direction vector a ₁ of the useful transmitter on the space defined by those of the interferers a ₂ . . . a _M such that:
• The scalar c[0170] _½−M then has the following expression: ${\mathrm{<figure-callout\; id="1"\; label="c"\; filenames="US20030103003A1-20030605-D00003.png,US20030103003A1-20030605-D00005.png"\; state="\{\{state\}\}">c</figure-callout>}}_{1/2-M}=\sqrt{\frac{{\underset{\_}{a}}_1^H{\Pi }_s\left(2,\dots ,M\right){\underset{\_}{a}}_1}{N}}$

  
• with Π[0171] _s(2, . . . , M)=A_2−M(A_2−M ^HA_2−M)⁻¹A_2−M ^H
• Where A[0172] _2−M=[a ₂ . . . a _m] and a _m ^H a _m=N for 1≦m≦M. In these conditions, the parameter c_½−M called the spatial correlation coefficient varies between 0 and 1. When c_½−M=0 the useful transmitter is orthogonal to the plane of the interferers and is therefore spatially decorrelated from the interferers. However, when c_½−M=1 the useful transmitter is in the plane of the interferers and becomes spatially consistent with the interferers. These spatial correlation configurations between the useful transmitter and the interferers will have a major influence on the performance characteristics of the direction-finding operations (FIG. 12).
Examples of Simulations with the Signals of the GSM
• In the simulations for this section, the signals s[0173] _m(t) are the GMSK modulations of the GSM. Since the signal sent has a constant modulus, the signal ŝ_m(t) is converted into {circumflex over ({circumflex over (s)})}_m(t) by the method (b). As above, δt=56 is chosen to be in the configuration of the signals TSC with f_e=600 kHz. The duration of the new reference signal s_m(t) will be lengthened to δt′=200 so that it corresponds to the approximate duration of a GSM burst. The signal is simulated on the circular array with f_ant=900 MHz. The direction-finding operations are performed in 1D and the incidence of the 1^st source is equal to θ₁=45 degrees.
• In FIG. 13, in the presence of only one source, the performance characteristics of the reference algorithms are compared with the signals d[0174] ₁(t) and {circumflex over ({circumflex over (s)})}₁(t) having respective durations of δt=56 and δt′=200. The Cramer Rao bounds associated with these integration times are plotted in solid lines. The direction-finding precision Δθ¹ _MES of the source is given as a function of its signal-to-noise ratio [S/B]₁.
• It can be seen that the method with the known signal d[0175] ₁(t) follows the Cramer Rao bound while the reference method with the signal {circumflex over ({circumflex over (s)})}₁(t) follows this bound beyond 0 dB. Thus, beyond 0 dB, the method with {circumflex over ({circumflex over (s)})}₁(t) is {square root}{square root over (200/56)}=1.88 times more precise than the method with d₁(t). Below 0 dB, the method with the reference sequence {circumflex over ({circumflex over (s)})}₁(t) is more precise with a gain of less than 1.88.
• FIG. 14 gives a schematic view of the results obtained by simulation of the case of a sequence d[0176] ₁(t) subjected to interference by a source without any synchronization signal. The coefficient of spatial correlation between the two sources is equal to c_½=0.9. The reference direction-finding algorithms are therefore single-user algorithms in using d₁(t) or {circumflex over ({circumflex over (s)})}₁(t) depending on the method. The curves represent the precision Δθ¹ _MES of the direction-finding operation for the 1^st source as a function of the signal-to-noise ratio [S/B] of the two sources.
• It can be seen that the reference methods using d[0177] ₁(t) or {circumflex over ({circumflex over (s)})}₁(t) have the same behavior in single-source operation: the method with d₁(t) follows its Cramer Rao bound and the method with the lengthened sequence {circumflex over ({circumflex over (s)})}₁(t) follows this bound beyond 0 dB in terms of signal-to-noise ratio. Since, from the theoretical viewpoint, the Cramer Rao bound with {circumflex over ({circumflex over (s)})}₁(t) is 1.88 times more precise than the one with d₁(t), this simulation clearly shows that the method with the sequence having a duration δt′ is more precise than the method with the sequence having a duration δt.
• The method described here above can be applied especially in a transmission system comprising a receiver consisting of N sensors connected to a device adapted to determining the directions of arrival of the different signals, such as a direction-finding device. The receiver comprises a processor equipped with a software program to implement the steps of the method, for all its variants of application. [0178]

Claims (10)

What is claimed is:

1 - A method of co-operative radio direction-finding in transmission in a system of radio direction-finding for a radiocommunications system comprising one or more transmission sources or transmitters, the transmitted signal comprising a reference sequence formed by a sequence of symbols {S_k}, the receiver comprising at least one array of several sensors coupled to a radio direction-finder, wherein the method comprises at least the following steps:

a) determining the reference signal or signals d_i(t) associated with a transmitter m from the reference symbols {S_k} and from the modulation defined by the format of the signals,

b) defining the instant of arrival t_k of one or more reference signals d_i(t),

c) from the signal x(t) received at the sensors, the instants of arrival t_k and the reference signal or signals d_i(t) derived from the step a), determining the value or values of incidence of the signals for a given transmitter.

2- A method according to claim 1, wherein the step a) is reiterated as many times as there are transmitters M₀ in the system.

3- A method according to claim 2 comprising a step for the separation of the transmitters M₀ through the estimation of the transfer matrix of the channel H^u for the transmission of all the useful signals and by the extraction, from this matrix, of the transfer matrices H_m ^u of the channel associated with a useful transmitter m in order to determine the corresponding observation vector x ^m(t).

4- A method according to claim 3, wherein the step c) is carried out by direction-finding by whitening, implemented on the observation vectors x ^m(t).

5- A method according to claim 3, wherein the step c) is carried out by direction-finding by projection.

6- A method according to one of the claims 1 to 3 wherein the direction-finding step uses the azimuth angle θ corresponding to the incidence of a source in the plane of the array and/or the elevation angle Δ defined as the incidence with respect to the perpendicular of the plane of the array.

7- A method according to one of the claims 1 to 6 wherein the step a) may be performed on a duration δt′>δt, where δt is the basic duration of the reference signal.

8- A method according to claim 7 wherein a spatial filter W_m is determined from the learning signal d_m(t) of a useful transmitter m in order to deduce the signal s_m (t) coming from the useful transmitter before applying a direction-finding step.

9- A use of the method according to one of the claims 1 to 8 in radio communications systems for GSM or UMTS type signals.

10- A direction-finding device comprising means adapted to the implementation of the method according to one of the claims 1 to 9.

Images